Quantitative Analysis of Simulated Illicit Street-Drug Samples Using Raman Spectroscopy and Partial Least Squares Regression

Modern drug laws require that a seized sample be characterized for both the illegal substances present and the quantity of each of those substances. The goal of this work was to develop a common approach to model development based on Raman spectroscopic analysis followed by partial least squares (PLS) regression that would allow us to obtain quantitative information from simulated street-drug samples. Each drug sample contained one drug surrogate—either isoxsuprine, norephedrine, benzocaine, or lidocaine—and up to 3 different cutting agents. All spectra were acquired on a homebuilt Raman instrument equipped with a rotating sample holder. The same steps were employed for developing separate models for each drug surrogate, including spectral preprocessing by Savitsky-Golay smoothing, differentiation, mean-centering, and autoscaling. PLS models were developed using 2 latent variables that yielded root mean square errors of calibration (RMSEC) values in the 3% range and root mean square error of prediction (RMSEP) values in the 4% range.